1 #ifndef __NET_SCHED_GENERIC_H 2 #define __NET_SCHED_GENERIC_H 3 4 #include <linux/netdevice.h> 5 #include <linux/types.h> 6 #include <linux/rcupdate.h> 7 #include <linux/pkt_sched.h> 8 #include <linux/pkt_cls.h> 9 #include <net/gen_stats.h> 10 #include <net/rtnetlink.h> 11 12 struct Qdisc_ops; 13 struct qdisc_walker; 14 struct tcf_walker; 15 struct module; 16 17 struct qdisc_rate_table { 18 struct tc_ratespec rate; 19 u32 data[256]; 20 struct qdisc_rate_table *next; 21 int refcnt; 22 }; 23 24 enum qdisc_state_t { 25 __QDISC_STATE_SCHED, 26 __QDISC_STATE_DEACTIVATED, 27 __QDISC_STATE_THROTTLED, 28 }; 29 30 /* 31 * following bits are only changed while qdisc lock is held 32 */ 33 enum qdisc___state_t { 34 __QDISC___STATE_RUNNING = 1, 35 }; 36 37 struct qdisc_size_table { 38 struct rcu_head rcu; 39 struct list_head list; 40 struct tc_sizespec szopts; 41 int refcnt; 42 u16 data[]; 43 }; 44 45 struct Qdisc { 46 int (*enqueue)(struct sk_buff *skb, struct Qdisc *dev); 47 struct sk_buff * (*dequeue)(struct Qdisc *dev); 48 unsigned int flags; 49 #define TCQ_F_BUILTIN 1 50 #define TCQ_F_INGRESS 2 51 #define TCQ_F_CAN_BYPASS 4 52 #define TCQ_F_MQROOT 8 53 #define TCQ_F_ONETXQUEUE 0x10 /* dequeue_skb() can assume all skbs are for 54 * q->dev_queue : It can test 55 * netif_xmit_frozen_or_stopped() before 56 * dequeueing next packet. 57 * Its true for MQ/MQPRIO slaves, or non 58 * multiqueue device. 59 */ 60 #define TCQ_F_WARN_NONWC (1 << 16) 61 int padded; 62 const struct Qdisc_ops *ops; 63 struct qdisc_size_table __rcu *stab; 64 struct list_head list; 65 u32 handle; 66 u32 parent; 67 atomic_t refcnt; 68 struct gnet_stats_rate_est rate_est; 69 int (*reshape_fail)(struct sk_buff *skb, 70 struct Qdisc *q); 71 72 void *u32_node; 73 74 /* This field is deprecated, but it is still used by CBQ 75 * and it will live until better solution will be invented. 76 */ 77 struct Qdisc *__parent; 78 struct netdev_queue *dev_queue; 79 struct Qdisc *next_sched; 80 81 struct sk_buff *gso_skb; 82 /* 83 * For performance sake on SMP, we put highly modified fields at the end 84 */ 85 unsigned long state; 86 struct sk_buff_head q; 87 struct gnet_stats_basic_packed bstats; 88 unsigned int __state; 89 struct gnet_stats_queue qstats; 90 struct rcu_head rcu_head; 91 spinlock_t busylock; 92 u32 limit; 93 }; 94 95 static inline bool qdisc_is_running(const struct Qdisc *qdisc) 96 { 97 return (qdisc->__state & __QDISC___STATE_RUNNING) ? true : false; 98 } 99 100 static inline bool qdisc_run_begin(struct Qdisc *qdisc) 101 { 102 if (qdisc_is_running(qdisc)) 103 return false; 104 qdisc->__state |= __QDISC___STATE_RUNNING; 105 return true; 106 } 107 108 static inline void qdisc_run_end(struct Qdisc *qdisc) 109 { 110 qdisc->__state &= ~__QDISC___STATE_RUNNING; 111 } 112 113 static inline bool qdisc_is_throttled(const struct Qdisc *qdisc) 114 { 115 return test_bit(__QDISC_STATE_THROTTLED, &qdisc->state) ? true : false; 116 } 117 118 static inline void qdisc_throttled(struct Qdisc *qdisc) 119 { 120 set_bit(__QDISC_STATE_THROTTLED, &qdisc->state); 121 } 122 123 static inline void qdisc_unthrottled(struct Qdisc *qdisc) 124 { 125 clear_bit(__QDISC_STATE_THROTTLED, &qdisc->state); 126 } 127 128 struct Qdisc_class_ops { 129 /* Child qdisc manipulation */ 130 struct netdev_queue * (*select_queue)(struct Qdisc *, struct tcmsg *); 131 int (*graft)(struct Qdisc *, unsigned long cl, 132 struct Qdisc *, struct Qdisc **); 133 struct Qdisc * (*leaf)(struct Qdisc *, unsigned long cl); 134 void (*qlen_notify)(struct Qdisc *, unsigned long); 135 136 /* Class manipulation routines */ 137 unsigned long (*get)(struct Qdisc *, u32 classid); 138 void (*put)(struct Qdisc *, unsigned long); 139 int (*change)(struct Qdisc *, u32, u32, 140 struct nlattr **, unsigned long *); 141 int (*delete)(struct Qdisc *, unsigned long); 142 void (*walk)(struct Qdisc *, struct qdisc_walker * arg); 143 144 /* Filter manipulation */ 145 struct tcf_proto ** (*tcf_chain)(struct Qdisc *, unsigned long); 146 unsigned long (*bind_tcf)(struct Qdisc *, unsigned long, 147 u32 classid); 148 void (*unbind_tcf)(struct Qdisc *, unsigned long); 149 150 /* rtnetlink specific */ 151 int (*dump)(struct Qdisc *, unsigned long, 152 struct sk_buff *skb, struct tcmsg*); 153 int (*dump_stats)(struct Qdisc *, unsigned long, 154 struct gnet_dump *); 155 }; 156 157 struct Qdisc_ops { 158 struct Qdisc_ops *next; 159 const struct Qdisc_class_ops *cl_ops; 160 char id[IFNAMSIZ]; 161 int priv_size; 162 163 int (*enqueue)(struct sk_buff *, struct Qdisc *); 164 struct sk_buff * (*dequeue)(struct Qdisc *); 165 struct sk_buff * (*peek)(struct Qdisc *); 166 unsigned int (*drop)(struct Qdisc *); 167 168 int (*init)(struct Qdisc *, struct nlattr *arg); 169 void (*reset)(struct Qdisc *); 170 void (*destroy)(struct Qdisc *); 171 int (*change)(struct Qdisc *, struct nlattr *arg); 172 void (*attach)(struct Qdisc *); 173 174 int (*dump)(struct Qdisc *, struct sk_buff *); 175 int (*dump_stats)(struct Qdisc *, struct gnet_dump *); 176 177 struct module *owner; 178 }; 179 180 181 struct tcf_result { 182 unsigned long class; 183 u32 classid; 184 }; 185 186 struct tcf_proto_ops { 187 struct tcf_proto_ops *next; 188 char kind[IFNAMSIZ]; 189 190 int (*classify)(struct sk_buff *, 191 const struct tcf_proto *, 192 struct tcf_result *); 193 int (*init)(struct tcf_proto*); 194 void (*destroy)(struct tcf_proto*); 195 196 unsigned long (*get)(struct tcf_proto*, u32 handle); 197 void (*put)(struct tcf_proto*, unsigned long); 198 int (*change)(struct sk_buff *, 199 struct tcf_proto*, unsigned long, 200 u32 handle, struct nlattr **, 201 unsigned long *); 202 int (*delete)(struct tcf_proto*, unsigned long); 203 void (*walk)(struct tcf_proto*, struct tcf_walker *arg); 204 205 /* rtnetlink specific */ 206 int (*dump)(struct tcf_proto*, unsigned long, 207 struct sk_buff *skb, struct tcmsg*); 208 209 struct module *owner; 210 }; 211 212 struct tcf_proto { 213 /* Fast access part */ 214 struct tcf_proto *next; 215 void *root; 216 int (*classify)(struct sk_buff *, 217 const struct tcf_proto *, 218 struct tcf_result *); 219 __be16 protocol; 220 221 /* All the rest */ 222 u32 prio; 223 u32 classid; 224 struct Qdisc *q; 225 void *data; 226 const struct tcf_proto_ops *ops; 227 }; 228 229 struct qdisc_skb_cb { 230 unsigned int pkt_len; 231 u16 slave_dev_queue_mapping; 232 u16 _pad; 233 unsigned char data[20]; 234 }; 235 236 static inline void qdisc_cb_private_validate(const struct sk_buff *skb, int sz) 237 { 238 struct qdisc_skb_cb *qcb; 239 240 BUILD_BUG_ON(sizeof(skb->cb) < offsetof(struct qdisc_skb_cb, data) + sz); 241 BUILD_BUG_ON(sizeof(qcb->data) < sz); 242 } 243 244 static inline int qdisc_qlen(const struct Qdisc *q) 245 { 246 return q->q.qlen; 247 } 248 249 static inline struct qdisc_skb_cb *qdisc_skb_cb(const struct sk_buff *skb) 250 { 251 return (struct qdisc_skb_cb *)skb->cb; 252 } 253 254 static inline spinlock_t *qdisc_lock(struct Qdisc *qdisc) 255 { 256 return &qdisc->q.lock; 257 } 258 259 static inline struct Qdisc *qdisc_root(const struct Qdisc *qdisc) 260 { 261 return qdisc->dev_queue->qdisc; 262 } 263 264 static inline struct Qdisc *qdisc_root_sleeping(const struct Qdisc *qdisc) 265 { 266 return qdisc->dev_queue->qdisc_sleeping; 267 } 268 269 /* The qdisc root lock is a mechanism by which to top level 270 * of a qdisc tree can be locked from any qdisc node in the 271 * forest. This allows changing the configuration of some 272 * aspect of the qdisc tree while blocking out asynchronous 273 * qdisc access in the packet processing paths. 274 * 275 * It is only legal to do this when the root will not change 276 * on us. Otherwise we'll potentially lock the wrong qdisc 277 * root. This is enforced by holding the RTNL semaphore, which 278 * all users of this lock accessor must do. 279 */ 280 static inline spinlock_t *qdisc_root_lock(const struct Qdisc *qdisc) 281 { 282 struct Qdisc *root = qdisc_root(qdisc); 283 284 ASSERT_RTNL(); 285 return qdisc_lock(root); 286 } 287 288 static inline spinlock_t *qdisc_root_sleeping_lock(const struct Qdisc *qdisc) 289 { 290 struct Qdisc *root = qdisc_root_sleeping(qdisc); 291 292 ASSERT_RTNL(); 293 return qdisc_lock(root); 294 } 295 296 static inline struct net_device *qdisc_dev(const struct Qdisc *qdisc) 297 { 298 return qdisc->dev_queue->dev; 299 } 300 301 static inline void sch_tree_lock(const struct Qdisc *q) 302 { 303 spin_lock_bh(qdisc_root_sleeping_lock(q)); 304 } 305 306 static inline void sch_tree_unlock(const struct Qdisc *q) 307 { 308 spin_unlock_bh(qdisc_root_sleeping_lock(q)); 309 } 310 311 #define tcf_tree_lock(tp) sch_tree_lock((tp)->q) 312 #define tcf_tree_unlock(tp) sch_tree_unlock((tp)->q) 313 314 extern struct Qdisc noop_qdisc; 315 extern struct Qdisc_ops noop_qdisc_ops; 316 extern struct Qdisc_ops pfifo_fast_ops; 317 extern struct Qdisc_ops mq_qdisc_ops; 318 319 struct Qdisc_class_common { 320 u32 classid; 321 struct hlist_node hnode; 322 }; 323 324 struct Qdisc_class_hash { 325 struct hlist_head *hash; 326 unsigned int hashsize; 327 unsigned int hashmask; 328 unsigned int hashelems; 329 }; 330 331 static inline unsigned int qdisc_class_hash(u32 id, u32 mask) 332 { 333 id ^= id >> 8; 334 id ^= id >> 4; 335 return id & mask; 336 } 337 338 static inline struct Qdisc_class_common * 339 qdisc_class_find(const struct Qdisc_class_hash *hash, u32 id) 340 { 341 struct Qdisc_class_common *cl; 342 struct hlist_node *n; 343 unsigned int h; 344 345 h = qdisc_class_hash(id, hash->hashmask); 346 hlist_for_each_entry(cl, n, &hash->hash[h], hnode) { 347 if (cl->classid == id) 348 return cl; 349 } 350 return NULL; 351 } 352 353 extern int qdisc_class_hash_init(struct Qdisc_class_hash *); 354 extern void qdisc_class_hash_insert(struct Qdisc_class_hash *, struct Qdisc_class_common *); 355 extern void qdisc_class_hash_remove(struct Qdisc_class_hash *, struct Qdisc_class_common *); 356 extern void qdisc_class_hash_grow(struct Qdisc *, struct Qdisc_class_hash *); 357 extern void qdisc_class_hash_destroy(struct Qdisc_class_hash *); 358 359 extern void dev_init_scheduler(struct net_device *dev); 360 extern void dev_shutdown(struct net_device *dev); 361 extern void dev_activate(struct net_device *dev); 362 extern void dev_deactivate(struct net_device *dev); 363 extern void dev_deactivate_many(struct list_head *head); 364 extern struct Qdisc *dev_graft_qdisc(struct netdev_queue *dev_queue, 365 struct Qdisc *qdisc); 366 extern void qdisc_reset(struct Qdisc *qdisc); 367 extern void qdisc_destroy(struct Qdisc *qdisc); 368 extern void qdisc_tree_decrease_qlen(struct Qdisc *qdisc, unsigned int n); 369 extern struct Qdisc *qdisc_alloc(struct netdev_queue *dev_queue, 370 struct Qdisc_ops *ops); 371 extern struct Qdisc *qdisc_create_dflt(struct netdev_queue *dev_queue, 372 struct Qdisc_ops *ops, u32 parentid); 373 extern void __qdisc_calculate_pkt_len(struct sk_buff *skb, 374 const struct qdisc_size_table *stab); 375 extern void tcf_destroy(struct tcf_proto *tp); 376 extern void tcf_destroy_chain(struct tcf_proto **fl); 377 378 /* Reset all TX qdiscs greater then index of a device. */ 379 static inline void qdisc_reset_all_tx_gt(struct net_device *dev, unsigned int i) 380 { 381 struct Qdisc *qdisc; 382 383 for (; i < dev->num_tx_queues; i++) { 384 qdisc = netdev_get_tx_queue(dev, i)->qdisc; 385 if (qdisc) { 386 spin_lock_bh(qdisc_lock(qdisc)); 387 qdisc_reset(qdisc); 388 spin_unlock_bh(qdisc_lock(qdisc)); 389 } 390 } 391 } 392 393 static inline void qdisc_reset_all_tx(struct net_device *dev) 394 { 395 qdisc_reset_all_tx_gt(dev, 0); 396 } 397 398 /* Are all TX queues of the device empty? */ 399 static inline bool qdisc_all_tx_empty(const struct net_device *dev) 400 { 401 unsigned int i; 402 for (i = 0; i < dev->num_tx_queues; i++) { 403 struct netdev_queue *txq = netdev_get_tx_queue(dev, i); 404 const struct Qdisc *q = txq->qdisc; 405 406 if (q->q.qlen) 407 return false; 408 } 409 return true; 410 } 411 412 /* Are any of the TX qdiscs changing? */ 413 static inline bool qdisc_tx_changing(const struct net_device *dev) 414 { 415 unsigned int i; 416 for (i = 0; i < dev->num_tx_queues; i++) { 417 struct netdev_queue *txq = netdev_get_tx_queue(dev, i); 418 if (txq->qdisc != txq->qdisc_sleeping) 419 return true; 420 } 421 return false; 422 } 423 424 /* Is the device using the noop qdisc on all queues? */ 425 static inline bool qdisc_tx_is_noop(const struct net_device *dev) 426 { 427 unsigned int i; 428 for (i = 0; i < dev->num_tx_queues; i++) { 429 struct netdev_queue *txq = netdev_get_tx_queue(dev, i); 430 if (txq->qdisc != &noop_qdisc) 431 return false; 432 } 433 return true; 434 } 435 436 static inline unsigned int qdisc_pkt_len(const struct sk_buff *skb) 437 { 438 return qdisc_skb_cb(skb)->pkt_len; 439 } 440 441 /* additional qdisc xmit flags (NET_XMIT_MASK in linux/netdevice.h) */ 442 enum net_xmit_qdisc_t { 443 __NET_XMIT_STOLEN = 0x00010000, 444 __NET_XMIT_BYPASS = 0x00020000, 445 }; 446 447 #ifdef CONFIG_NET_CLS_ACT 448 #define net_xmit_drop_count(e) ((e) & __NET_XMIT_STOLEN ? 0 : 1) 449 #else 450 #define net_xmit_drop_count(e) (1) 451 #endif 452 453 static inline void qdisc_calculate_pkt_len(struct sk_buff *skb, 454 const struct Qdisc *sch) 455 { 456 #ifdef CONFIG_NET_SCHED 457 struct qdisc_size_table *stab = rcu_dereference_bh(sch->stab); 458 459 if (stab) 460 __qdisc_calculate_pkt_len(skb, stab); 461 #endif 462 } 463 464 static inline int qdisc_enqueue(struct sk_buff *skb, struct Qdisc *sch) 465 { 466 qdisc_calculate_pkt_len(skb, sch); 467 return sch->enqueue(skb, sch); 468 } 469 470 static inline int qdisc_enqueue_root(struct sk_buff *skb, struct Qdisc *sch) 471 { 472 qdisc_skb_cb(skb)->pkt_len = skb->len; 473 return qdisc_enqueue(skb, sch) & NET_XMIT_MASK; 474 } 475 476 477 static inline void bstats_update(struct gnet_stats_basic_packed *bstats, 478 const struct sk_buff *skb) 479 { 480 bstats->bytes += qdisc_pkt_len(skb); 481 bstats->packets += skb_is_gso(skb) ? skb_shinfo(skb)->gso_segs : 1; 482 } 483 484 static inline void qdisc_bstats_update(struct Qdisc *sch, 485 const struct sk_buff *skb) 486 { 487 bstats_update(&sch->bstats, skb); 488 } 489 490 static inline int __qdisc_enqueue_tail(struct sk_buff *skb, struct Qdisc *sch, 491 struct sk_buff_head *list) 492 { 493 __skb_queue_tail(list, skb); 494 sch->qstats.backlog += qdisc_pkt_len(skb); 495 496 return NET_XMIT_SUCCESS; 497 } 498 499 static inline int qdisc_enqueue_tail(struct sk_buff *skb, struct Qdisc *sch) 500 { 501 return __qdisc_enqueue_tail(skb, sch, &sch->q); 502 } 503 504 static inline struct sk_buff *__qdisc_dequeue_head(struct Qdisc *sch, 505 struct sk_buff_head *list) 506 { 507 struct sk_buff *skb = __skb_dequeue(list); 508 509 if (likely(skb != NULL)) { 510 sch->qstats.backlog -= qdisc_pkt_len(skb); 511 qdisc_bstats_update(sch, skb); 512 } 513 514 return skb; 515 } 516 517 static inline struct sk_buff *qdisc_dequeue_head(struct Qdisc *sch) 518 { 519 return __qdisc_dequeue_head(sch, &sch->q); 520 } 521 522 static inline unsigned int __qdisc_queue_drop_head(struct Qdisc *sch, 523 struct sk_buff_head *list) 524 { 525 struct sk_buff *skb = __skb_dequeue(list); 526 527 if (likely(skb != NULL)) { 528 unsigned int len = qdisc_pkt_len(skb); 529 sch->qstats.backlog -= len; 530 kfree_skb(skb); 531 return len; 532 } 533 534 return 0; 535 } 536 537 static inline unsigned int qdisc_queue_drop_head(struct Qdisc *sch) 538 { 539 return __qdisc_queue_drop_head(sch, &sch->q); 540 } 541 542 static inline struct sk_buff *__qdisc_dequeue_tail(struct Qdisc *sch, 543 struct sk_buff_head *list) 544 { 545 struct sk_buff *skb = __skb_dequeue_tail(list); 546 547 if (likely(skb != NULL)) 548 sch->qstats.backlog -= qdisc_pkt_len(skb); 549 550 return skb; 551 } 552 553 static inline struct sk_buff *qdisc_dequeue_tail(struct Qdisc *sch) 554 { 555 return __qdisc_dequeue_tail(sch, &sch->q); 556 } 557 558 static inline struct sk_buff *qdisc_peek_head(struct Qdisc *sch) 559 { 560 return skb_peek(&sch->q); 561 } 562 563 /* generic pseudo peek method for non-work-conserving qdisc */ 564 static inline struct sk_buff *qdisc_peek_dequeued(struct Qdisc *sch) 565 { 566 /* we can reuse ->gso_skb because peek isn't called for root qdiscs */ 567 if (!sch->gso_skb) { 568 sch->gso_skb = sch->dequeue(sch); 569 if (sch->gso_skb) 570 /* it's still part of the queue */ 571 sch->q.qlen++; 572 } 573 574 return sch->gso_skb; 575 } 576 577 /* use instead of qdisc->dequeue() for all qdiscs queried with ->peek() */ 578 static inline struct sk_buff *qdisc_dequeue_peeked(struct Qdisc *sch) 579 { 580 struct sk_buff *skb = sch->gso_skb; 581 582 if (skb) { 583 sch->gso_skb = NULL; 584 sch->q.qlen--; 585 } else { 586 skb = sch->dequeue(sch); 587 } 588 589 return skb; 590 } 591 592 static inline void __qdisc_reset_queue(struct Qdisc *sch, 593 struct sk_buff_head *list) 594 { 595 /* 596 * We do not know the backlog in bytes of this list, it 597 * is up to the caller to correct it 598 */ 599 __skb_queue_purge(list); 600 } 601 602 static inline void qdisc_reset_queue(struct Qdisc *sch) 603 { 604 __qdisc_reset_queue(sch, &sch->q); 605 sch->qstats.backlog = 0; 606 } 607 608 static inline unsigned int __qdisc_queue_drop(struct Qdisc *sch, 609 struct sk_buff_head *list) 610 { 611 struct sk_buff *skb = __qdisc_dequeue_tail(sch, list); 612 613 if (likely(skb != NULL)) { 614 unsigned int len = qdisc_pkt_len(skb); 615 kfree_skb(skb); 616 return len; 617 } 618 619 return 0; 620 } 621 622 static inline unsigned int qdisc_queue_drop(struct Qdisc *sch) 623 { 624 return __qdisc_queue_drop(sch, &sch->q); 625 } 626 627 static inline int qdisc_drop(struct sk_buff *skb, struct Qdisc *sch) 628 { 629 kfree_skb(skb); 630 sch->qstats.drops++; 631 632 return NET_XMIT_DROP; 633 } 634 635 static inline int qdisc_reshape_fail(struct sk_buff *skb, struct Qdisc *sch) 636 { 637 sch->qstats.drops++; 638 639 #ifdef CONFIG_NET_CLS_ACT 640 if (sch->reshape_fail == NULL || sch->reshape_fail(skb, sch)) 641 goto drop; 642 643 return NET_XMIT_SUCCESS; 644 645 drop: 646 #endif 647 kfree_skb(skb); 648 return NET_XMIT_DROP; 649 } 650 651 /* Length to Time (L2T) lookup in a qdisc_rate_table, to determine how 652 long it will take to send a packet given its size. 653 */ 654 static inline u32 qdisc_l2t(struct qdisc_rate_table* rtab, unsigned int pktlen) 655 { 656 int slot = pktlen + rtab->rate.cell_align + rtab->rate.overhead; 657 if (slot < 0) 658 slot = 0; 659 slot >>= rtab->rate.cell_log; 660 if (slot > 255) 661 return rtab->data[255]*(slot >> 8) + rtab->data[slot & 0xFF]; 662 return rtab->data[slot]; 663 } 664 665 #ifdef CONFIG_NET_CLS_ACT 666 static inline struct sk_buff *skb_act_clone(struct sk_buff *skb, gfp_t gfp_mask, 667 int action) 668 { 669 struct sk_buff *n; 670 671 n = skb_clone(skb, gfp_mask); 672 673 if (n) { 674 n->tc_verd = SET_TC_VERD(n->tc_verd, 0); 675 n->tc_verd = CLR_TC_OK2MUNGE(n->tc_verd); 676 n->tc_verd = CLR_TC_MUNGED(n->tc_verd); 677 } 678 return n; 679 } 680 #endif 681 682 #endif 683